One hundred fifty-three community-dwelling older people aged 65 years and older took part in this international, multicentre, single-blinded, two-group randomized trial (Fig. 1). Study sites were located in Germany (Cologne), Spain (Valencia) and Australia (Sydney). The trial was conducted between January and October 2014. A study protocol describing the applied system and methodologies in more detail is available elsewhere .
Older people were included if they met the following eligibility criteria: (1) aged 65 years and older, (2) living in the community, (3) able to walk 20 m without a walking aid, (4) able to watch television TV with or without their glasses from 3 m distance, and (5) have enough space for system use (3.5 m2). The exclusion criteria were: (1) insufficient language skills to understand the study procedures, (2) cognitive impairment (Mini-Cog: 1–2 recalled words and abnormal clock drawing test) , and (3) medical conditions precluding participation in a regular exercise program (i.e., uncontrolled hypertension, severe neurological disorder, acute cancer, psychiatric disorder, acute infection).
Randomization and blinding
Following baseline assessments, eligible participants were randomised by permuted block-randomisation (ratio 1:1) using a unique computer-generated random number for identification. Participants who lived in the same household were treated as one unit and randomised into the same block. Research staff performing the assessments was experienced, trained and blinded to group allocation. Participants were reminded not to talk about their user experience to avoid unblinding.
All participants gave written informed consent prior to inclusion. Ethical approval was obtained by the ethics committees of the German Sport University Cologne (24.09.2013), the Polytechnic University of Valencia (19.12.2013), and the Human Research Ethics Committee of the University of New South Wales (reference number HC12316, 19.12.2013). After baseline assessment (see outcome measures described below), participants were formally entered into the study and randomised to intervention or control groups (Fig. 1). Participants in both groups received an evidence-based educational booklet about general health and fall prevention . Control group participants did not receive any additional intervention and were encouraged to follow their habitual exercise routines if applicable. Participants in the intervention group were instructed in the use of the iStoppFalls program in their home, including tailored and targeted (personalised) balance and muscle strength exercises (exergames), and a test battery for the assessment of individual fall risk. Trained research staff installed the iStoppFalls system components including a personal computer (Shuttle Barebone Slim-PC), a Google TV set top box (STB) by Sony, a Microsoft Kinect (3D depth sensor), a Senior Mobility Monitor (SMM) by Philips (3D accelerometer, barometer) , and a Nexus 7 Android tablet in the homes of intervention group participants. Participants were instructed how to control the movements of a virtual avatar on the TV screen and how to navigate through the system with gestures, voice control or use of the tablet. Two weeks after system installation, a second home visit ensured correct and safe system use as well as progression of training. Phone support was available throughout the 16-week intervention and additional home visits were offered if required.
Intervention group participants conducted a 16-week exercise program based on best practice recommendations for exercise to prevent falls in older people by Sherrington et al. [15, 16]. The recommended training dose consisted of at least three balance sessions of about 40 min each (including each of the exergames) and at least three muscle strength sessions of about 15 to 20 min each (including all strength exercises) per week. A sheet of ‘Exercise Safety Guidelines’ was given to each participant by an experienced researcher. Adherence was monitored automatically by the iStoppFalls system. Participants received individual training and assessment reminders through the tablet computer and by an unblinded research assistant if required. All aggregated data were transmitted to a knowledge-based system (server) which allowed participants to continuously monitor their fall risk and results via the STB and tablet computer at home.
Balance exercises were based on the Weight-bearing Exercise for Better Balance (WEBB) program (www.webb.org.au). Three balance exergames ‘Bumble Bee Park’, ‘Hills & Skills’, and ‘Balance Bistro’ for walking, weight shifting, knee bending, and/or stepping in different directions were specifically developed for the iStoppFalls project. Additionally, cognitive tasks targeting semantic and working memory (e.g., remembering objects) were added once a participant reached higher exergame levels (dual-tasking). Progression was achieved by reducing upper limb support, narrowing the base of support, adjusting speed of movement, increasing gaming duration, and proceeding to a higher difficulty level.
Strength exercises for the lower extremities including knee extension, knee flexion, hip abduction, calf raises, and toe raises were based on the strength exercise component of the Otago exercise program . Between 2 and 3 sets of 10 to 15 repetitions and rest periods of 1 min were recommended. Progression was achieved by increasing the number of repetitions, the number of sets, and the difficulty level (e.g., by using 1–3 kg ankle cuff weights).
The whole sample was assessed at baseline (0 weeks) and at the end of the intervention period (16 weeks). A self-report questionnaire was used to collect information on socio-demographic characteristics and medical history. Anthropometrics were assessed as part of the baseline assessments. Falls frequency and adverse events were monitored with monthly diaries for 6 months. Participants were contacted by phone when the diaries were not returned.
Primary outcome measures
The short version of the Physiological Profile Assessment (PPA) estimated individual fall risk based on five sensorimotor tests: contrast sensitivity (Melbourne edge test (MET), peripheral sensation (proprioception), balance (sway when standing on medium-density foam with eyes open), lower extremity muscle strength (knee extension), and hand reaction time (HRT) . The European Quality of Life 5 Dimensions (EQ-5D) questionnaire was used to assess health status in five dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression (www.euroqol.org/eq-5d-products/eq-5d-5l.html).
Secondary outcome measures
The 12-item World Health Organization Disability Assessment Schedule (WHODAS) 2.0 was used to assess general health (understanding and communicating, mobility, self-care, interpersonal interactions, household and work activities, and participation in society) . The 9-item Patient Health Questionnaire (PHQ-9) was used to assess the severity of depression . Participants’ concerns about falling for 10 daily activities were investigated by the shortened Iconographical Falls Efficacy Scale (Icon-FES) . The Incidental and Planned Activity Questionnaire (IPEQ) in Spain and Australia , and the Physical Activity Questionnaire for the population aged 50 years and older (PAQ-50+) in Germany  were applied to retrospectively assess physical activity patterns. The SMM was used in the intervention group participants to detect walking distance and sit-to-stand transfers during daily life activities, using previously defined algorithms . Participants were asked to wear the SMM during waking hours. The generated peak power was calculated for each detected sit-to-stand transfer, with an expected detection sensitivity of 89 % .
In addition to the PPA, tests for coordinated stability and maximal balance range were assessed as measures of dynamic balance. Tests for static balance (e.g., tandem stance), walking speed over 4 m, and five times chair stand performance from the Short Physical Performance Battery (SPPB)  were also administered. The timed up and go test (TUG) was used as a combination of basic functionality, physical mobility, and dynamic balance [26–28]. Steady-state walking speed was measured over a 10 m distance (plus 2 m for acceleration and 2 m for deceleration) with a stop watch . Dual-tasking ability was assessed by asking participants to count backwards by three starting from a random 3-digit number while walking over a 10 m distance . Furthermore, four specifically developed sensor-based physical tests were performed: 1) balance (bipedal, semi-tandem, near-tandem, and tandem stance); 2) arm reaction time (hitting two randomly flashing lights on a virtual table by lifting one arm), 3) stepping reaction time (step on two randomly flashing lights on a virtual floor); and 4) five times sit-to-stand (stand up and sit down).
The Trail Making Test (TMT) was performed as a measure of executive function (divided attention), processing/motor speed, and mental flexibility, in which participants had to connect numbers and lines in order [31, 32]. Cognitive control was assessed by a computer-based Victoria Stroop Test (VST) using the Psychology Experiment Building Language (PEBL) software version 0.13 (http://pebl.sourceforge.net/). In the VST, participants had to maintain a goal in mind and supress habitual responses to correctly identify coloured dots and words . The Digit Symbol Coding Test (DSC) required participants to copy symbols to investigate processing speed . Working memory, attention, and concentration was measured by the Digit Span Backward (DSB) . In this test, participants had to repeat numbers in the reverse order . A computer-based Attention Network Test (ANT) where participants had to determine whether a central arrow points to the left or right was used on PEBL to quantify the processing efficiency within three attentional networks: alerting, orienting and executive attention .
Technology use measures
Usability and enjoyment was assessed by the 10-item System Usability Scale (SUS)  and the 8-item Physical Activity Enjoyment Scale (PACES) [36, 37]. A score of 0–24 on the PACES corresponds to a lower overall enjoyment, and a score of 24–48 corresponds to a higher enjoyment. Overall usability measured by the SUS ranges from 0 “worst imaginable” to 100 “best imaginable”. The Dynamic Acceptance Model for the Re-evaluation of Technologies (DART) was used for the analysis and evaluation of user acceptance of products or services . It comprises items related to appeal, consistency, operation, speed, language and usability on a 6-point Likert scale (6 indicates “very important”/”totally fulfilled”).
Statistical analyses and sample size
Sample size calculations based on PPA as the primary outcome estimated a sample size of 52 participants (f = 0.40, alpha 5 %, power 80 %) . With an anticipated dropout rate of 15 %, the recruitment aimed for 60 participants per site. For this study the intention-to-treat (ITT) method including post-assessment data from exercise dropouts was applied. Data on feasibility was analysed using descriptive techniques. Student t-tests for continuous variables with normal distribution, chi-square test for nominal data, and Mann–Whitney U test for ordinal or continuous data without normal distribution were used to determine differences between the intervention and control group at baseline. Repeated measures ANOVA was used to determine the intervention effect on outcome measures at follow-up. Three-way (comparing 3 groups) and two-way (comparing high-adherence group to control group) repeated mesuares ANOVAs were used to perform subgroup analyses for grouping variables of interest (exercise adherence and fall risk based on median split). The two-sided alpha level was 5 %. Analyses were performed with SPSS version 23 for Windows (SPSS, Inc., Chicago, IL).